Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on September 25, 2013 3:14 am wrote:
> That looks like an i7 overclocked to 4.8GHz. Pretty much all i7-4770 results are 32-bit
> and in the 3500-3600 range. After a bit of browsing I found a 64-bit i7 result:
>
> http://browser.primatelabs.com/geekbench3/compare/81817?baseline=43547
>
> That suggests a 3.9GHz i7 is 2.8x faster than A7 on single threaded integer with a 3x clock advantage.



But that's comparing two chips with wildly different clock rates. Memory latency in ns is fixed regardless of a CPU's clock rate, and there is limited ability to help the latency (in ns) of L1/L2 cache, as well.

Basically at a low clock rate, main memory is 1/3 as far away in terms of cycles, and cache is ~1/2 as far away in terms of cycles. That's why low clocked CPUs like Itanium always look better than they really are in terms of IPC. The i7 is sitting around twiddling its thumbs waiting for data a whole lot more, simply because of its 3x higher clock rate.

If it were possible to take the A7 and clock it at 3.9 GHz, adjusting the L1/L2 access as necessary, it would not perform anything like the i7. I don't know what it would score, but you seem to be assuming linear scaling when that isn't what you'll get. And of course, it couldn't clock 3x as fast, it would need a complete redesign, much of which would further reduce its performance at 3.9 GHz.

Plus all this assumes geekbench is actually useful, and I'm highly skeptical of that. Possibly it has some value to compare phones, but comparing a phone to a high end desktop? Switching to some other benchmark that chooses a different set of mostly useless microbenchmarks to average over could produce very different results.